Sign in with
Sign up | Sign in
Your question

7870 or 660

Tags:
  • Graphics Cards
  • GPUs
  • Intel i5
  • Graphics
  • Product
Last response: in Graphics & Displays
Share
November 12, 2012 4:26:34 AM

Hello everyone,I'm gettin the i5-3570k and I'm wondering what gpu to buy, the 7870 or the 660, which one is a better buy

Thanks!

More about : 7870 660

a b U Graphics card
November 12, 2012 4:36:27 AM

The performance of both is very similar. Personally I would go for the 660.
m
0
l
a c 143 U Graphics card
November 12, 2012 4:48:19 AM

@Rezn
Did you mean GTX 660ti or GTX 660?

GTX 660ti is about 6% faster than HD 7870, but right now you can get HD 7870 for as low as $210.
m
0
l
Related resources
a b U Graphics card
November 12, 2012 5:02:53 AM

if the 3 free games means anything to you and you prefer high MSAA over FXAA for whatever reason, the 7870. otherwise, the 660 will perform close enough with it.
m
0
l
a c 143 U Graphics card
November 12, 2012 5:06:14 AM

Actually, I'd get the HD 7870 even if I hate AMD, the 3 games are "WTF" deal :D 
m
0
l
a b U Graphics card
November 12, 2012 6:56:30 AM

that 3 free games deal isnt available everywhere in the world so it doesnt count.
both are great cards but i would go for whichever is cheaper
m
0
l
a b U Graphics card
November 12, 2012 8:01:18 AM

Like Ironslice says, near enough the same. I'd go with a GTX660 if it was me.
m
0
l
November 12, 2012 11:20:31 PM

Get a 7870 and screw the nuts off it, you will be keeping up and beating the Ti once overclocked the 660 will get no where near those speeds.
m
0
l
November 13, 2012 7:03:03 AM

First of all the 7870 doesn't come with 3 free games... It only comes with one, which is Far Cry 3. :p  The 3 game deal is only eligible if you get two 7870's (for crossfire) or get a 7950-7970 I think?

Secondly: I asked the same question about a week ago here: http://www.tomshardware.com/forum/376671-33-help-choose...

There were a lot of votes for the 7870, but I ended up getting a 660 anyway. Both cards really are very similar in performance, especially the OC 660's vs the GHZ edition AMD cards. But I think it really comes down to your personal preferences and what you want to use these cards for.

I own a gigabyte 660 2GB OC card. Performance is pretty damn good overall, and t runs just about everything I throw at it with max settings. On an i5-2500k running at the stock speed. Most games get around 50-75FPS at 1200 resolution with about 4xAA and 8xAF. I have never gone above 75FPS because this monitor has a maximum refresh rate of only 75Hz unfortunately.

Actually this monitor only has a maximum resolution of 1680x1050 (it has a 16:10 aspect ratio), but I am able to game at 1920x1200 by using the resolution down sampling option in the Nvidia display driver options. Games look so clean and jaggy free when you run a 1920x1200 down sampled image on a 1680x1050 resolution monitor. It is really one of the nicer features available in the driver settings.

As for overclocking this card, well there isn't much room for overhead to be honest. I can only really clock this card maybe just 65MHz increase on the GPU clock and 100Mhz increase on the memory clock. But this card was already overclocked to begin with, so I guess it should be expected. It offers some pretty awesome results without having to push high levels of AA and FA.

PhysX is nice to have in some games like XCom and Borderlands 2, but it is not something I would miss if they were removed.

The card using a very low power draw for the performance it gets, which is always a nice thing, and has a dynamic GPU core that reduces power usage when not being pushed very hard. Heat temps stay nice and chilly on this model. Really I am getting lower temps at peak performance now than when I was using just the onboard video on the CPU.


As far as I can tell the 7870 is a better overclockers card because of its higher memory bandwidth (256mb vs 192mb) which might make it a better choice if you want to push higher frame rates. Also people claim that it handles higher levels of AA better at higher resolutions than the 660. The latest AMD drivers do support resolution down sampling as far as I know, which is an awesome feature if your card can support it. But this card also draws more power overall as far as the reviews go.

I think if you want a little more in the way of features without worrying about maximum performance, go with Nvidia. If you want higher performance in general with overclocking possibilities go with AMD.
m
0
l
a b U Graphics card
November 13, 2012 7:12:31 AM

You mean bus width, not bandwidth :-) And 256-bit vs 192-bit. It's actually irrelevant anyway though, pretty much just a myth. Funnily enough, bandwidth actually is the real determining factor in memory performance, measured in gigs per second, and it's almost the same for both options.

Bandwidth is the product of bus width and the memory clock speed. You could think of it like a road, where bus width is the number of lanes of traffic, while memory clock speed is the speed that traffic is moving. Every lane is full of cars moving at the same speed. To double the number of cars reaching the end of the road per second, you can either double the number of lanes, or you can double the speed the cars are moving. Two methods, same result. Similarly, to achieve an increased memory bandwidth, you could either increase the memory clocks or increase the bus width. Again, two methods, same result.

Now the GTX660 and GTX660 Ti use memory clocked at 1500MHz (6000MHz effective) vs 1200MHz from the 7800-series cards. That 25% clock speed gain goes a long way to closing the bandwidth gap. The 7800 cards still have the bandwidth lead, but only by ~5%.

Anyway at the end of the day, even if I hadn't explained bandwidth, I could make the same point by simply referring you to the benchmarks, where you'd see just how capable a card the GTX660 is. All the ROPs, MHz and GB/s in the world don't count for a thing next to frames per second. These numbers are a means to an ends, the end result being framerates. Using these other numbers for comparison is akin to basing a processor purchase on clock speed alone - it's an uninformed and inaccurate way of doing things. Hopefully people will become more aware of this in the future :-)
m
0
l
November 13, 2012 9:26:25 AM

sam_p_lay said:
You mean bus width, not bandwidth :-) And 256-bit vs 192-bit. It's actually irrelevant anyway though, pretty much just a myth. Funnily enough, bandwidth actually is the real determining factor in memory performance, measured in gigs per second, and it's almost the same for both options.

Bandwidth is the product of bus width and the memory clock speed. You could think of it like a road, where bus width is the number of lanes of traffic, while memory clock speed is the speed that traffic is moving. Every lane is full of cars moving at the same speed. To double the number of cars reaching the end of the road per second, you can either double the number of lanes, or you can double the speed the cars are moving. Two methods, same result. Similarly, to achieve an increased memory bandwidth, you could either increase the memory clocks or increase the bus width. Again, two methods, same result.

Now the GTX660 and GTX660 Ti use memory clocked at 1500MHz (6000MHz effective) vs 1200MHz from the 7800-series cards. That 25% clock speed gain goes a long way to closing the bandwidth gap. The 7800 cards still have the bandwidth lead, but only by ~5%.

Anyway at the end of the day, even if I hadn't explained bandwidth, I could make the same point by simply referring you to the benchmarks, where you'd see just how capable a card the GTX660 is. All the ROPs, MHz and GB/s in the world don't count for a thing next to frames per second. These numbers are a means to an ends, the end result being framerates. Using these other numbers for comparison is akin to basing a processor purchase on clock speed alone - it's an uninformed and inaccurate way of doing things. Hopefully people will become more aware of this in the future :-)


Well, I am not the most knowledgeable on the subject of buswidth/ bandwidth/ clock speeds, I was just going by what little I knew... or at least thought I knew :p  . But thanks for the corrections, that's some good information there.

As for benchmarks, I have tested quite a few games in my library on this card using fraps. Here's a few of my own conclusions:

Borderlands 2:
Resolution: 1920x1200 (using down sampling)
all settings max
16x Anisotropic Filtering
PhysX: medium
Framerate setting: unlimited

The framerate hovers around 60-65 at this resolution setting, but can dip down to the 30's in pretty heavy firefights.

XCom:
Resolution: 1920x1200 (using down sampling)
all settings max
Vsync enabled
Framerate smoothing off

Surprisingly gets around 75fps here. It does have minor dips in areas with heavier action, but overall very smooth.

Alan Wake:
Resolution: 1920x1200 (using down sampling)
all settings max
8x AntiAliasing
16x Anisotropic Filtering

gets about 50-52fps in most large open areas, but can bog down to about 30-40 when heavy shader effects are in use and multiple enemies are on screen. Though I have never seen this game drop below 30 at this resolution. I generally lower the AA and FA a bit to get better framerates in this one.

GTAIV:
Resolution: 1920x1200 (using down sampling)
All settings maxed out
16x anisotropic filtering
vehicle density, view distance, LOD: all at 100%

This is probably one of the most taxing games I have in my library I would say. Either that or the game is poorly optimized. But this game can bring even the highest end PC's down to their knees, it it runs at around 40-45fps max at these settings, but the framerate can drop down to 25-27 when the traffic picks up to incredibly large numbers.

Deus Ex: Human Revolution:
Resolution: 1920x1200 (using down sampling)
DX11 mode
all settings maxed out
16x anisotropic
AA: MLAA
Shadows: Soft

Runs at a smooth 75fps in most areas, but can drop down to at least 45fps at its lowest (that I have noticed) during heavier firefights and enemy encounters.

But anyway, it's a bit hard for me to judge how high most of these games actually get in framerate as my monitor does not go over 75hz (75fps), so I cannot give an accurate numbers above that cap. But in just about all cases I have been getting 40-60fps at 1920x1200 with all settings maxed out. Aside from GTA 4 which can dip down to the high 20's, I haven't seen any other game in my library drop below 30fps. Every valve game I have tried, including games like Portal 2, Dota2, L4D2, all stay above 75FPS no matter what I throw at them.

Yeah overall it is a really competent card for the price.
m
0
l
a b U Graphics card
November 13, 2012 9:29:36 AM

Glad to spread the knowledge :-) Thanks for those benchmarks, what kind of monitor are you using there? Borderlands 2 actually runs flawlessly on my old 4870!
m
0
l
a b U Graphics card
November 13, 2012 9:30:57 AM

Also, does anyone understand the framerate setting in BL2? Is it like some kind of v-sync? And what's going on with 22-62fps smoothed?
m
0
l
November 13, 2012 9:56:22 AM

sam_p_lay said:
Glad to spread the knowledge :-) Thanks for those benchmarks, what kind of monitor are you using there? Borderlands 2 actually runs flawlessly on my old 4870!


Umm... looking at the hardware profile, its a Dell E228WFP. The monitor is 16:10 and has a maximum resolution at 1680x1050 at 75hz. I salvaged it from an old Dell PC that somebody wanted to throw away, because I was to cheap to buy a new one for this PC :p  . The contrast is decent considering how old he monitor is and it works well enough for gaming. But as I have said before, the Nvidia display drivers has an option to create down sampled resolutions which I use to play games at 1900x1200 on this screen.

sam_p_lay said:
Also, does anyone understand the framerate setting in BL2? Is it like some kind of v-sync? And what's going on with 22-62fps smoothed?


Yeah I am not too sure about that myself. I assume that the 22-62fps smoothed tries to find a middle ground between your highest and lowest framerates in the game to try and get a smoother performance. But I could be wrong. :p 
m
0
l
a b U Graphics card
November 13, 2012 10:06:46 AM

I think they've just complicated things there... there may be genuine benefit to how smooth and/or responsive the gameplay feels, but it's not very clear how it works or what to choose! Surely nobody wants to restrict their framerate? And 22-62fps just seems like some strange numbers to choose... I think I just set it as 60fps to correspond with v-sync, seems to run fine anyway so I leave it :-)
m
0
l
!